“Why did we do it this way again?”
The decision is six months old, the person who made it has left, and the thread that settled it is still somewhere.
The data layer between your tools and any AI.
Connect your tools once. Ask anything from Claude, ChatGPT or your own agents, always with the source.
Toute forme revient
Data takes the shape of the system that holds it.
Here, it finds one again.
Art by Mada Erzinoskaïa
Your conversations, your documents, your files.
Your tools, your data, available to any AI.
What Bucket does, in four moves.
Connect once. Ask in plain language. Control who sees what. Keep it working with any AI.
Connect your sources
Plug your tools in once. It runs in the background.
Slack, Notion, GitHub, Google Drive, Discord. PDFs, contracts, decks. Bucket ingests them passively, so your team keeps working exactly as before.
Ask anything
The exact answer, from the exact moment it was said or written.
Dated, sourced, and attributed to the right person. Not a summary. A graph shows you where every answer comes from.
Fine-grained permissions
Everyone sees exactly what they should. Nothing else.
Permissions are respected at the source. Decide who reaches which data, workspace by workspace, and every flow stays in its lane.
Works with any AI
Your knowledge stays yours. Plug it into any AI.
Claude, ChatGPT, Cursor, or whatever your team moves to next. No lock-in, exportable at any time. Your knowledge follows you.
Built on MCPConnect Bucket to any AI in one paste.
One MCP endpoint. Add it to Claude, ChatGPT, Cursor or your terminal. Nothing else to install.
Add a custom connector
Connect Bucket to your AI. Paste the endpoint from your workspace settings.
Name
Bucket
Shown in your connector list.Remote MCP server URL
https://mcp.bckt.dev/acme-corp/sse
The HTTPS address where the server accepts MCP requests.Advanced settings
bkt_9f2a41c7e60b
OAuth client ID (optional)••••••••••••3d1e
OAuth client secret (optional)CancelAdd
Works with
Claude
ChatGPT
CursorClaude Code
Codex
Your data lake, and who reads it.
Connect a tool once. Each bucket takes what it needs. Everyone reads their own through the Relay.
Acme Corp
Everything your tools hold lands here, once. This is your organisation’s data lake. Buckets are the slices of it your teams work from, and everyone reads their own from the AI they already use.
Connected tools
One connection per tool, shared by every bucket here. Disconnecting removes its data from the whole organisation.
SlackconnectedDisconnect- Engineering3
- Go-to-market2
NotionconnectedDisconnect- Engineering2
GitHubconnectedDisconnect- Engineering2
Google DriveNot connectedConnect
Who uses it
A bucket sees the tools it was given, plus its own private files. Open one to change what it takes.
- Engineering7 sources
- Go-to-market2 sources
- Adam’s bucketno sources yet
- FinancePrivate
Relay
Your Relay
One endpoint of your own, covering every bucket you belong to. Point the AI you already use at it, and it reads what you can read. Nothing more.
Speaks the Model Context Protocol. Not available yet.
The moments every team runs into.
The same memory, wherever it’s missing.
“Has anyone answered this client before?”
The answer was written three times, by three people, in three different tools.
“Can you send me that doc again?”
The most asked question in any company, and the one that costs everyone the most time.
“I need to onboard the new hire.”
Three months rebuilding what the team already knows, by interrupting the people who know it.
Questions we get asked.
Why not just use the connectors built into ChatGPT or Claude?
Those connect one person’s account, with that person’s access. Everyone on the team has to set it up again, in their own corner, and no one has a view of who can reach what. They also search one source at a time, so a question whose answer spans a Slack thread and a Confluence page comes back incomplete. Bucket connects once, for the organization, with permissions.
How is this different from an AI agent platform?
Those platforms are where you go to work. You build assistants inside their interface, and your team has to move there. Bucket has no interface for your team. Your people keep the AI they already use (Claude, ChatGPT, Cursor), and it simply knows your company.
Why store the data instead of querying the tools live?
Because you can’t search what you don’t hold. Live APIs search by keyword, not by meaning, and they return whole threads and whole pages, which floods the model’s context with noise. Bucket indexes your content, then returns the few passages that actually answer the question. That selection is the product.
What happens if we leave?
Full export, permanent deletion, on request. MCP is an open standard, so nothing about your setup is specific to us.
What if our tool isn’t supported?
If it has an API, we build the connector with you. Scoped upfront, delivered as a service.
Who can see what?
Permissions are respected at the source. If someone can’t open a document in Google Drive, they don’t get it through Bucket either. Workspaces let you separate teams on top of that.
Give your AI the data it needs.
Connect your tools once. Works with whatever comes next.
Talk to sales